Blog

What 10 or 30 hours of file transcription a month costs on Otter.ai, Rev, Sonix, HappyScribe and ScribeToAny — cost per hour, free plans and limits (Sept 2026).

An 11MB worker bundle and cold starts meant a 3.5s TTFB. How edge HTML caching, bundle splitting and deferred hydration got us to a 95+ Lighthouse score.

Workers can't run a multi-minute GPU job or Modal's gRPC SDK, so we call Modal over HTTP and get results back by signed webhook. The design and its sharp edges.

Captions add non-speech sounds for viewers who can't hear the audio; subtitles translate dialogue for those who don't speak the language. Which you need.

What Word Error Rate means, how transcription accuracy is really measured, why even humans don't hit 100%, and what moves AI accuracy up or down.

A reference to transcript and subtitle formats — SRT, VTT, TXT, TSV, CSV, JSON, ASS, SSA, SMI and TTML — with real examples and when to use each.